test: rewrite unit tests with testify/goleak; coverage 100% (#10)

[millerjp]

Summary

Replaces the fork's internal-package smoke test with a comprehensive external-package suite and raises the CI coverage gate to the AxonOps 95% target. Coverage on syncmap.go jumps from 36.1% to 100.0%.

Closes #10.

What's in this PR

Two commits, bisectable:

1. test: rewrite unit tests in external package with testify/goleak (#10) — replaces syncmap_smoke_test.go with syncmap_test.go, adds testify v1.11.1 and goleak v1.3.0 to go.mod, creates go.sum.
2. ci: raise coverage threshold to 95% (#10) — flips threshold=35.0 → 95.0 in .github/workflows/ci.yml, removes the TODO(#11) comment.

Test structure

• package syncmap_test (black-box).
• TestMain wires goleak.VerifyTestMain(m) as the single package-wide leak hook (per-test VerifyNone(t) was dropped because t.Parallel() siblings would detect each other's goroutines).
• 15 top-level TestX functions, 33 subtests, 48 total t.Parallel() calls.
• Every public method: Load, Store, LoadOrStore, LoadAndDelete, Delete, Range, Len, Map, Keys, Items — each with happy path, missing-key, zero-value distinction, and edge cases.
• Four concurrency tests:
  • TestConcurrentWritersReaders — 16 writers × 8 readers × 1000 ops, writer-id bounds asserted on loaded values.
  • TestLoadOrStoreContention — 100 goroutines, single key, exactly one "stored" winner (via atomic.Int32).
  • TestRangeDuringWrites — writers cycle through the whole keyspace concurrently with Range; subset invariant asserted per sync.Map contract.
  • TestDeleteDuringRange — deleter concurrent with Range; no panic; observed (k, v) pairs satisfy k == v.
• Every concurrency test stable over go test -race -count=10.

Coverage — per method

Load               100.0%
Store              100.0%
LoadOrStore        100.0%
LoadAndDelete      100.0%
Delete             100.0%
Range              100.0%
Len                100.0%
Map                100.0%
Keys               100.0%
Items              100.0%
total              100.0%

Agent gates

• test-analyst (pre-code): scenario plan
• test-writer: produced the initial file
• test-analyst (post-code): confirmed every scenario landed, branches are genuinely exercised (!ok in Load, !l in LoadAndDelete, both Range callback paths)
• code-reviewer: approved with two IMPORTANT tightenings — reader keyspace variation in TestConcurrentWritersReaders, whole-keyspace writer loop in TestRangeDuringWrites. Both applied.
• security-reviewer: no blockers; race-clean, no over-strong assertions, no panic paths in test code
• devops (on the threshold flip): approved
• go-quality: all checks pass (the earlier "FAIL" on go mod tidy + diff was the pre-commit dirty state, resolved on commit)
• commit-message-reviewer: both commit messages pass

Local validation

$ make check
… (fmt-check vet lint tidy-check test test-bdd coverage security)
coverage: 100.0% of statements
No vulnerabilities found.
(exit 0)

$ go test -race -count=10 -run 'TestConcurrent|TestLoadOrStoreContention|TestRangeDuringWrites|TestDeleteDuringRange' ./...
ok  github.com/axonops/syncmap  1.609s

Test plan

• All CI jobs green (especially the newly-strict coverage gate)
• test matrix stable on both ubuntu-latest and macos-latest
• Coverage artefact downloadable and reports 100.0%